Presentation attack detection for face recognition using light field camera.

Area of Science:

• Computer Vision
• Biometrics
• Security Systems

Background:

• Face recognition systems are increasingly vulnerable to sophisticated presentation attacks (spoof attacks).
• Existing face presentation attack detection (PAD) schemes lack a universally superior method due to evolving attack sophistication.
• There is a critical need for advanced PAD techniques to enhance biometric security.

Purpose of the Study:

• To introduce a novel perspective for face presentation attack detection (PAD) using light field camera (LFC) technology.
• To explore the potential of focus variations captured by LFCs for distinguishing genuine faces from spoof attacks.
• To develop and evaluate a new PAD scheme based on LFC-rendered depth images.

Main Methods:

• Utilized a light field camera (LFC) to capture multi-depth images of faces in a single shot.
• Developed a novel PAD approach by analyzing the variation of focus across multiple depth images rendered by the LFC.
• Created a new face artefact database using LFC, comprising 80 subjects and simulating photo print and electronic screen attacks.

Main Results:

• The proposed PAD scheme demonstrated outstanding performance in detecting presentation attacks.
• Experiments on the light field face artefact database showed superior efficacy compared to established state-of-the-art PAD schemes.
• The analysis of focus variations proved effective in revealing sophisticated spoofing attempts.

Conclusions:

• Light field cameras offer a unique and effective approach for face presentation attack detection.
• Analyzing focus variations from LFC-rendered images is a promising method for enhancing biometric security against spoofing.
• The proposed LFC-based PAD scheme represents a significant advancement in combating sophisticated face presentation attacks.